Re: [PATCH v6 2/3] CPU hotplug, stop-machine: Plug race-window that leads to "IPI-to-offline-CPU"

["Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com>]

On 05/23/2014 08:51 PM, Peter Zijlstra wrote:
> On Fri, May 23, 2014 at 03:42:20PM +0530, Srivatsa S. Bhat wrote:
>>        Re-enable interrupts                          Re-enable interrupts
>>
>>                                                      The pending IPI is noted
>>                                                      immediately, but alas,
>>                                                      the CPU is offline at
>>                                                      this point.
>>
> 
> So wasn't clear_local_APIC() (and the arch function __cpu_disable() in
> general) wipe all pending interrup state?
> 

Apparently not. Atleast powerpc explicitly re-enables only the IPIs.. 

A snippet from xics_migrate_irqs_away() in powerpc, which is called
in the cpu-disable phase:


/* Interrupts are disabled. */
void xics_migrate_irqs_away(void)
{
	[...]
	/* Reject any interrupt that was queued to us... */
	icp_ops->set_priority(0);

	[...]
	/* Allow IPIs again... */
	icp_ops->set_priority(DEFAULT_PRIORITY);

	for_each_irq_desc(virq, desc) {
		[...]
		/* We only need to migrate enabled IRQS */
		if (!desc->action)
			continue;

		[...]

		/* We need to get IPIs still. */
		if (irq == XICS_IPI || irq == XICS_IRQ_SPURIOUS)
			continue;
		[...]

}

But that said, it should have cleared any pending IPIs and allowed
only *new* IPIs (for whatever reason). Atleast the warning I hit
indicates that that didn't happen, and the old IPI really was taken
later.

Ben, Paul, any thoughts on this?

Regards,
Srivatsa S. Bhat